moollm（moollm 搜索）

MOOLLM

*"Many-Voiced Object-Oriented LLM — the system that explains itself."*

What Is It?

The moollm skill is the spirit and constitution of MOOLLM itself. It's the top-level help agent that can:

• Explain what MOOLLM is
• Answer "what can I do?"
• Navigate users to relevant skills
• Articulate the philosophy
• Show the constitution
• Recommend approaches for tasks

When confused, invoke this skill.

The Core Ideas

Many-Voiced

MOOLLM doesn't use a single LLM perspective. It simulates multiple agents debating within a single call:

• Committees of personas with opposing views
• Deliberation forced by Robert's Rules
• Evaluation by independent assessors
• The debate produces wisdom, not statistics

Filesystem as World Model

Directories are rooms. Files are objects:

examples/adventure-4/
├── pub/              # A room
│   ├── ROOM.yml      # Room properties
│   ├── pie-table.yml # An object
│   └── cat-cave/     # Nested room
├── characters/       # Metaphysical room
└── ADVENTURE.yml     # Game state

Play-Learn-Lift

The methodology:

1. PLAY — Explore freely, try things, fail safely
2. LEARN — Notice patterns, document what works
3. LIFT — Share as reusable skills

Skills as Prototypes

Skills are documented capabilities that can be:

• Instantiated into specific contexts
• Composed with other skills
• Inherited (multiple inheritance)
• Evolved through play

Protocol

When invoked, this skill should:

1. Assess what the user needs
2. If lost → provide orientation
3. If asking "what can I do?" → show relevant capabilities
4. If asking about philosophy → explain core concepts
5. If asking about skills → navigate to skill system
6. Always be helpful, welcoming, and clear

Constitutional File Map (kernel/)

• kernel/README.md — index and navigation for kernel docs
• kernel/constitution-core.md — core constitution and invariants
• kernel/constitution-template.md — template for new constitutions
• kernel/ARCHITECTURE.md — system architecture overview
• kernel/context-assembly-protocol.md — context assembly rules
• kernel/memory-management-protocol.md — memory, limits, persistence
• kernel/event-logging-protocol.md — logging and provenance
• kernel/tool-calling-protocol.md — tool usage contract
• kernel/self-healing-protocol.md — recovery and repair behaviors
• kernel/DIRECTORY-AS-OBJECT.md — directory as object model
• kernel/SELFISH-COM-IMPLEMENTATION.md — SELF lineage in practice
• kernel/INTEREST-GATES.yml — attention gating rules
• kernel/NAMING.yml — top-level naming policy
• kernel/naming/NAMING.yml — detailed naming rules
• kernel/naming/NAMING-K-LINES.yml — K-line naming standards
• kernel/naming/NAMING-CONSTELLATIONS.yml — constellation naming
• kernel/naming/NAMING-COMPILATION.yml — compiled name patterns
• kernel/naming/NAMING-PATH-VARIABLES.yml — path variable rules
• kernel/naming/NAMING-RELATIONSHIPS.yml — relationship naming
• kernel/naming/URLS.yml — URL conventions
• kernel/drivers/README.md — driver index
• kernel/drivers/cursor.yml — Cursor-specific driver
• kernel/drivers/generic.yml — baseline driver
• kernel/drivers/custom.yml — site-specific overrides
• kernel/drivers/claude-code.yml — Claude Code driver
• kernel/drivers/antigravity.yml — experimental driver

Local Runtime Files (.moollm/)

These are gitignored runtime files for session state, scratch, and logs.

• .moollm/working-set.yml — current focus and active files
• .moollm/hot.yml — priority hints
• .moollm/cold.yml — cold-start state
• .moollm/startup.yml — startup context
• .moollm/output.md — append-only output log
• .moollm/session-log.md — append-only session log
• .moollm/bootstrap-probe.yml — bootstrap probes and checks

Cursor Boot Optimization (cursor-mirror)

Use cursor-mirror to inspect boot state, reduce context bloat, and verify setup.

Commands (need composer ID):

• tree — list sessions/composers
• status — quick health check
• tail --limit 50 — recent messages
• timeline <composer> — full event sequence
• thinking <composer> — reasoning blocks
• tools <composer> — tool call history
• grep <pattern> — search transcripts

Use this to:

• confirm which bootstrap ran
• locate missing context
• triage performance issues
• audit tool-call provenance

Plan: MOOLLM Linter, Mirror, Compiler

Phase 0 — Meta Inhale (Planning Only)

1. Inventory the repository structure and file types.
2. Define categories: essential, primary, secondary, hidden.
3. Establish ignore rules (build output, caches, temp files, editor artifacts).
4. Identify canonical docs to preserve: top-level README.md, designs/, docs/.
5. Map root entry points to their authoritative specs (skills, kernels, protocols).
6. Reverse-engineer the root README.md into a structured spec:

- Purpose - Audience - Sections - Source references - Link strategy

1. Define a report format for validation output:

- Structural warnings - Missing declarations - Naming consistency - Cross-link integrity

1. Define "what to hide" for human-facing outputs (noise, duplication, low-signal).
2. Define "what to surface" for quick navigation (skills index, starting points).
3. Define sister-script scope: lint, mirror, compile.
4. Write a sister-script design brief (no code yet).

Inputs

• User questions about MOOLLM
• Requests for help or navigation
• Philosophical inquiries

Outputs

• Clear explanations
• Skill recommendations
• Navigation guidance
• Philosophy articulation

Dovetails With

• skill/ — How skills work
• k-lines/ — K-lines and naming
• play-learn-lift/ — The methodology
• kernel/constitution-core.md — The constitution

Protocol Symbol

MOOLLM-HELP

Invoke when: User is confused, lost, or wants to understand the system.

See: PROTOCOLS.yml